leakaudit: Detect and Audit Train/Test Leakage from Near-Duplicate Images

Detects near-duplicate images across dataset splits using perceptual hashing, reports the resulting train/validation/test contamination, and produces a corrected, leak-free split assignment. Intended for machine learning researchers who need to verify that image classification splits do not share near-duplicate samples across partitions before reporting model metrics.

Version: 0.1.0
Depends: R (≥ 4.1)
Imports: magick
Suggests: testthat (≥ 3.0.0), knitr, rmarkdown
Published: 2026-07-29
DOI: 10.32614/CRAN.package.leakaudit (may not be active yet)
Author: Kartik Patel [aut, cre], Samruddhi Amol Shah [aut], BitandByte [cph]
Maintainer: Kartik Patel <kartikpatel.id at gmail.com>
BugReports: https://github.com/anakincodex/leakaudit/issues
License: MIT + file LICENSE
URL: https://github.com/anakincodex/leakaudit
NeedsCompilation: no
Materials: README, NEWS
CRAN checks: leakaudit results

Documentation:

Reference manual: leakaudit.html , leakaudit.pdf
Vignettes: Getting started with leakaudit (source, R code)

Downloads:

Package source: leakaudit_0.1.0.tar.gz
Windows binaries: r-devel: not available, r-release: not available, r-oldrel: not available
macOS binaries: r-release (arm64): leakaudit_0.1.0.tgz, r-oldrel (arm64): leakaudit_0.1.0.tgz, r-release (x86_64): leakaudit_0.1.0.tgz, r-oldrel (x86_64): leakaudit_0.1.0.tgz

Linking:

Please use the canonical form https://CRAN.R-project.org/package=leakaudit to link to this page.